Anyone running DeepStack 2022.01.1?

[Corvus85]

So I'm back to the problem I was having before I reinstalled DS.

For some reason, BI is still reporting '404' errors in cancelled alerts. It only seems to be happening at night.




The custom model folder is correct:




And my AI settings in the Trigger tab haven't changed at all in weeks. Still using the 'dark' model for my night profile.



Any ideas?

 

[wittaj]

As I mentioned in Post #31 - Even though BI let's you point where the custom models are, many have found that if you do not put them in the default location that BI assigns, that it is problematic, so put them in C:\DeepStack\MyModels

It is probably something in the coding that is looking for a file structure. I had trouble originally until I put it in C:\DeepStack\MyModels

 

[Corvus85]

As I mentioned in Post #31 - Even though BI let's you point where the custom models are, many have found that if you do not put them in the default location that BI assigns, that it is problematic, so put them in C:\DeepStack\MyModels

It is probably something in the coding that is looking for a file structure. I had trouble originally until I put it in C:\DeepStack\MyModels

Seems that did it. Strange... It never needed that to work fine before I reinstalled DS :-\

But here's what kind of performance I'm getting out of it. It's giving me 'Error 100', and the mainstream isn't kicking in until well after DS recognized me.



is there a setting I need to change somewhere?

 

[Corvus85]

Anyone?

 

[MikeLud1]

Seems that did it. Strange... It never needed that to work fine before I reinstalled DS :-\

But here's what kind of performance I'm getting out of it. It's giving me 'Error 100', and the mainstream isn't kicking in until well after DS recognized me.

View attachment 117455

is there a setting I need to change somewhere?

Does this happen when you use just the combined model? The dark model is harder on the system if you are using the CPU version of DeepStack.

 

[Corvus85]

Does this happen when you use just the combined model? The dark model is harder on the system if you are using the CPU version of DeepStack.

Well during the day, I have it set to use just the combined model and it works fine.

I understand that the dark model is harder on the system, but it was working fine a few days ago :-\

 

[105437]

Well during the day, I have it set to use just the combined model and it works fine.

I understand that the dark model is harder on the system, but it was working fine a few days ago :-\

You need to be willing to try things and troubleshoot by using process of elimination. Go back to a standard, default model setup for all profiles. This will quickly rule out the dark model if the problem persists.

 

[Corvus85]

Go back to a standard, default model setup for all profiles. This will quickly rule out the dark model if the problem persists.

I'll try it again tonight.

 

[105437]

I'll try it again tonight.

Remove it from the MyModels folder.

 

[sebastiantombs]

I've been tracking the number of instances of python being held in memory on my system for a few weeks. It has ranged from eight to as many as 20, apparently depending on how many cameras have recently had images analyzed. Since updating to 2022.01.1 it has been either five or ten. Additionally, before the update a typical python instance occupied between three and four GB of RAM. Since the update that amount has increased to four to five GB. My system has 32GB, but with ten instances of python in memory occupying a total of 30GB of RAM it's not a pretty picture. I'll be dropping back to the previous version of DS later this afternoon.

This may be specific to my system but it's driving me nuts.

 

[MikeLud1]

I've been tracking the number of instances of python being held in memory on my system for a few weeks. It has ranged from eight to as many as 20, apparently depending on how many cameras have recently had images analyzed. Since updating to 2022.01.1 it has been either five or ten. Additionally, before the update a typical python instance occupied between three and four GB of RAM. Since the update that amount has increased to four to five GB. My system has 32GB, but with ten instances of python in memory occupying a total of 30GB of RAM it's not a pretty picture. I'll be dropping back to the previous version of DS later this afternoon.

This may be specific to my system but it's driving me nuts.

How many NVIDIA Containers do you have, before I updated to 2022.01.1 I was starting to have memory issue and saw 6 NVIDIA Containers and similar issues. What I did is uninstall everything including CUDA and started from a clean slate and now I have the below 2 NVIDIA Containers and a Python for each model I am using .

 

[sebastiantombs]

I'll have a look at that. I'm running two instances of DS as configured in BI.

 

[MikeLud1]

I'll have a look at that. I'm running two instances of DS as configured in BI.

I did some testing to see if it helps running more than one instances of DS and found it does not help the speed. I tested with two Python scripts loading DeepStack as fast as the scripts can and saw no benefit of having two instances of DS. I will run the test again over the weekend and post some screenshot show the speed.

 

[sebastiantombs]

I think it depends on the specific card when it comes to instances.

With two instances on a GTX970 I get consistent sub 100ms times during the day with the combined model and sub 400ms times at night using dark and combined. With one instance I get regular timeouts. Keep in mind that I have three cameras that trigger basically simultaneously with vehicle traffic. One is an approach view, one a broadside and one departing view.

 

[MikeLud1]

I think it depends on the specific card when it comes to instances.

I am using a RTX 3060 so the video card should have plenty of power to see if running more than one instances will help with response times.

 

[sebastiantombs]

@MikeLud1 What driver version are you using for the 3060?

 

[MikeLud1]

@MikeLud1 What driver version are you using for the 3060?

See below, I found any newer just did not work right.

 

[MikeLud1]

@MikeLud1 What driver version are you using for the 3060?

Below are the CUDA version I am using.

 

[sebastiantombs]

Thanks Mike. Checking on things now. There are three NVidia containers running, two NvidiaDisplay containers (two cards), and three NVidia share.exe running.

 

[Fubduck]

It's been almost a week now and the Jetson Nano is working fine with 2022-01-1 and BI 5.5.4.5. I am not running any custom models. I am only feeding it 2 LPR cameras during the day nothing at night. Response times are low to mid 200ms.